Reinforcement Learning-Based Multi-Domain Network Slice Composition with Topology Aggregation
Publication Date
1-1-2026
Document Type
Conference Proceeding
Publication Title
2026 International Conference on Computing Networking and Communications Icnc 2026
DOI
10.1109/ICNC68183.2026.11416842
First Page
207
Last Page
212
Abstract
The establishment of network slices across multiple network domains requires orchestrations across domains, which is challenging due to the heterogeneity of domain-specific features and limited visibility into each domain's state information. Topology Aggregation (TA) enables domains to provide aggregated network representations in cross-domain collaborations. We propose a reinforcement learning (RL)-based framework to minimize the deployment cost of network slice provisioning across multi-domain networks. This framework takes the slice service request and aggregated network state representations encoded by Graph Neural Networks (GNNs) as input, and yields a composition plan which consists of the construction of the network slice and the placement of service functions. The simulation results show that the proposed framework enables more cost-effective slice deployment compared to a heuristic approach.
Keywords
network slicing, reinforcement learning, service function chain, topology aggregation
Department
Computer Science; Aviation and Technology
Recommended Citation
Congzhou Li, Genya Ishigaki, Zhouxiang Wu, Divya Khanure, Riti Gour, and Jason P. Jue. "Reinforcement Learning-Based Multi-Domain Network Slice Composition with Topology Aggregation" 2026 International Conference on Computing Networking and Communications Icnc 2026 (2026): 207-212. https://doi.org/10.1109/ICNC68183.2026.11416842
